Scholarlink

This study investigates the prediction of symptom escalation in oncology by categorizing 12 self-reported symptoms into physical (e.g., nausea, fatigue, pain) and mental (e.g., feeling blue, trouble thinking) groups. Symptom data were grouped into 3- to 7-day intervals to balance the highly imbalanced dataset, where 84% of entries showed no escalation. The models—Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU)—were trained on 80% of the data and evaluated on the remaining 20%. Results showed that 3-day intervals achieved the highest accuracy across all models. For physical symptoms, CNN performed the best, achieving an accuracy of 79.2%, precision of 84.1%, recall of 78.8%, and an F1 score of 81.4% at 3-day intervals. GRU showed superior results for mental symptoms with an accuracy of 77.2%, precision of 71.6%, recall of 62.2%, and an F1 score of 66.6% at 3-day intervals. Performance metrics declined with longer intervals due to reduced temporal resolution and fewer training samples, though CNN and GRU retained greater stability in their predictions. These findings highlight the effectiveness of categorizing symptoms into physical and mental groups for tailored predictions. In addition, our results underscore the potential of deep learning models in providing actionable insights into symptom trajectories. Integrating these predictive models into clinical workflows can enable proactive symptom management, allowing for timely interventions and improved patient outcomes during chemotherapy.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning Approaches for Prediction of Mental and Physical Deterioration During Chemotherapy

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Joseph Finkelstein,
  • Aref Smiley,
  • Christina Echeverria,
  • Kathi Mooney

摘要

This study investigates the prediction of symptom escalation in oncology by categorizing 12 self-reported symptoms into physical (e.g., nausea, fatigue, pain) and mental (e.g., feeling blue, trouble thinking) groups. Symptom data were grouped into 3- to 7-day intervals to balance the highly imbalanced dataset, where 84% of entries showed no escalation. The models—Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU)—were trained on 80% of the data and evaluated on the remaining 20%. Results showed that 3-day intervals achieved the highest accuracy across all models. For physical symptoms, CNN performed the best, achieving an accuracy of 79.2%, precision of 84.1%, recall of 78.8%, and an F1 score of 81.4% at 3-day intervals. GRU showed superior results for mental symptoms with an accuracy of 77.2%, precision of 71.6%, recall of 62.2%, and an F1 score of 66.6% at 3-day intervals. Performance metrics declined with longer intervals due to reduced temporal resolution and fewer training samples, though CNN and GRU retained greater stability in their predictions. These findings highlight the effectiveness of categorizing symptoms into physical and mental groups for tailored predictions. In addition, our results underscore the potential of deep learning models in providing actionable insights into symptom trajectories. Integrating these predictive models into clinical workflows can enable proactive symptom management, allowing for timely interventions and improved patient outcomes during chemotherapy.